Prospective Study of the Diagnostic Accuracy of the Simplify D-dimer Assay for Pulmonary Embolism in Emergency Department Patients: Criterion Standard
Imaging was performed according to a flow diagram posted in the ED. The primary pulmonary vascular imaging study was CT angiography of the chest and venography of the legs, performed and interpreted as we have previously described. Patients who were allergic to iodinated contrast or had a serum creatinine measurement > 1.5 mg/dL underwent ventilation-perfusion scintillation lung scanning, interpreted by board-certified radiologists with specialty training in nuclear medicine in accordance with Prospective Investigation of Pulmonary Embolism Diagnosis study criteria. Lower-extremity venous ultrasound was ordered at the discretion of the attending emergency physician. Radiologists who interpreted images were unaware of the of the d-dimer result. Patients with negative d-dimer results did not necessarily undergo pulmonary vascular imaging. For all patients, the criterion standard was the result of 90-day follow-up, using a structured combination of telephone and medical record follow-up as we have previously described.
The diagnosis of PE included patient-verified and medical record-verified evidence of a new diagnosis of PE or deep venous thrombosis (DVT) requiring treatment, or death from PE within 90 days after testing with the d-dimer assay. The details of the methodology of defining PE or DVT discovered on follow-up have been pub-lished. Briefly, this required that two independent physicians agree that the patient had image- or autopsy-proven PE or DVT diagnosed within the follow-up period, and that treatment (either anticoagulation, vena caval interruption, or both) was initiated in living patients. Patients with either PE or DVT or both were considered PE positive (PE + ). Patients who were alive and had no diagnosis of PE or DVT after 90 days were considered PE negative (PE— ).
Data are presented in accordance with the Standards for Reporting of Diagnostic Accuracy initiative. Primary analyses are from 2 X 2 diagnostic tables, and all analyses were performed using statistical software (version 2.2.4; StatsDirect; Cheshire, UK). Diagnostic indexes and 95% confidence intervals (CIs) were calculated by the exact (Clopper-Pearson) method. To examine for spectrum bias, we used first-order regression analysis and Pearson correlation coefficient (R) to examine the plots of sensitivity, specificity, and negative likelihood ratio as a function of prevalence of PE. An unpaired t test was applied to the slope of the regression lines to test for a significant correlation with p < 0.05 considered significant.